In the realm of multimedia forensics, source identification is one of the most prominent issues. Identifying the origin of the media is used in detecting forged media as well as tracing the questioned media to its source imaging device. Consequently, law enforcement worldwide employs these techniques as evidence in their investigations including courts of law. In the last decade, this topic has received significant interest from the research community resulting in several techniques. One of the distinctive noise patterns used to identify digital media to its originating device is Photo Response Non-Uniformity (PRNU). Research in recent years has concentrated on using noise patterns in conjunction with machine learning technologies. This thesis aims to propose a novel automated smartphone identification system for forensic investigation. The system utilizes the PRNU noise, which is extracted from the still images taken directly from the smartphone with or without post-processing. The fingerprints are fed into the proposed deep learning model to detect and identify the manufacturer and model of the smartphone that the image originated from. The system achieved an omni-model (device specific) identification accuracy of 97.60% on UNIFI dataset. Furthermore, the system achieves intra-model identification accuracy of 90.17%, 95.73%, and 95.8% on the top three smartphone brands, Apple, Huawei, and Samsung respectively. In addition, the system utilizes Extreme Learning Machines (ELM) concept to expand the models without the costly fully re training the deep learning models.